Application of classic and soft computing for modeling yield and environmental final impact in vegetable production (a case study: transplanting onion in Isfahan province, Iran)

This study aimed to develop a precision model between inputs and yield, and also between inputs (indirect emissions) and environmental final index (EFI) in onion farms through regression models (classic computing) and artificial intelligence models (soft computing). Required data were collected through direct measurement and questionnaire. To this end, 85 and 70 questionnaires were distributed among onion farmers in Fereydan and Falavarjan regions (Isfahan province, center of Iran), respectively. In the Fereydan region, the total energy input, onion yield, and water use efficiency (WUE) were obtained as 239496 MJ.ha⁻¹, 97658 kg.ha⁻¹, and 9.08 kg.m⁻³, respectively, while for Falavarjan region, these were obtained as 232221 MJ.ha⁻¹, 94485 kg.ha⁻¹, and 10.8 kg m⁻³, respectively. Electricity and diesel fuel were the most widely used inputs in the study areas. Based on the results related to the environmental indices, EFI was obtained as 547.38 and 363.54 pPt.t⁻¹ for Fereydan and Falavarjan regions, respectively. The contribution of direct (such as CO₂ and NH₃) and indirect emissions (especially electricity) to the total EFI was 74 and 26% in Fereydan and 63 and 37% in Falavarjan region, respectively. Results related to the Cobb–Douglas regression model (CDR) showed that the effects of seed, manure, and labor on the onion yield were significant at 1% level of confidence. However, despite meeting the regression assumptions, the CDR model has predicted the yield and EFI with lower accuracy and higher error compared to artificial neural network models (ANNs), multi-layer perceptron (MLP), and adaptive neuro-fuzzy inference system (ANFIS). Soft computing (artificial intelligence) modeling showed that the ANFIS model (Grid Partitioning (GP)) has higher computational speed an lower error compared to multi-layer perceptron (MLP) models. Therefore, the comparison of the best GP and MLP models showed that the root-mean-square-error (RMSE) was obtained as 10.649 and 52.321 kg.ha⁻¹ for yield and 25.08 and 40.94 pPt.ha⁻¹ for EFI, respectively.